Electron Energy Loss Spectroscopy Investigation into Symmetry in

Aug 2, 2016 - We present a combined scanning transmission electron microscopy–electron energy loss spectroscopy (STEM–EELS) investigation into the...
2 downloads 12 Views 4MB Size
Electron Energy Loss Spectroscopy Investigation into Symmetry in Gold Trimer and Tetramer Plasmonic Nanoparticle Structures Steven J. Barrow,†,∥ Sean M. Collins,‡,∥ David Rossouw,§ Alison M. Funston,⊥ Gianluigi A. Botton,§ Paul A. Midgley,‡ and Paul Mulvaney*,† †

School of Chemistry and Bio21 Institute, University of Melbourne, Parkville, Victoria 3010, Australia Department of Materials Science and Metallurgy, University of Cambridge, 27 Charles Babbage Road, Cambridge CB3 0FS, United Kingdom § Materials Science and Engineering, McMaster University, Hamilton, Ontario L8S 4L8, Canada ⊥ Chemistry Department, Monash University, Clayton, Victoria 3800, Australia ‡

S Supporting Information *

ABSTRACT: We present a combined scanning transmission electron microscopy−electron energy loss spectroscopy (STEM−EELS) investigation into the mode symmetries of plasmonic nanoparticle trimer and tetramer structures. We obtain nanometer-resolved energy loss spectra for both trimer and tetramer structures and compare these to boundary element method simulations. We show that EELS, in conjunction with eigenmode simulations, offers a complete characterization of the individual superstructures, and we trace the evolution of both optically dark and bright modes and identify multipolar mode contributions. We then apply this technique to tetramer structures that exhibit an expanded range of mode symmetries for two-dimensional and three-dimensional selfassembled geometries. These findings provide a comprehensive experimental account of the available photonic states in self-assembled nanoparticle clusters. KEYWORDS: STEM−EELS, surface plasmon, nanoparticle assemblies, non-negative matrix factorization, electron energy loss spectroscopy

P

Plasmon coupling in small nanoparticle assemblies has been likened to that of molecular orbital theory, where plasmons from interacting particles can hybridize giving rise to “bonding” and “antibonding” states.10 Discrete nanoparticle assemblies can thus be viewed as “plasmonic molecules”, where the plasmon modes of individual nanoparticles behave collectively as a result of plasmon hybridization. Furthermore, the overall plasmonic behavior of discrete nanoparticle assemblies has been shown to be largely dependent on structure symmetry. Irreducible representations of nanoparticle assemblies, derived from group theory, have been demonstrated to provide a solid basis for understanding the plasmon modes of these systems.11,12 However, the experimental observation of the plasmonic properties of such nanoparticle structures has been achieved predominantly with optical techniques.7,11−16 Modes

lanar arrangements of self-assembled nanoparticles have generated a substantial research following due to the unique surface plasmon phenomena they support. Circularly arranged nanostructures are of particular interest due to the accommodation of Fano-like resonances which can lead to enhanced localized electric fields in the vicinity of these structures.1−3 Such field enhancements are useful in the amplification of Raman signals for chemical sensing,4,5 and it has been proposed that circular two-dimensional (2D) nanoassemblies could be used for plasmon-based optical switching.6 Moreover, three-dimensional (3D) nanostructures are predicted to be useful meta-materials with negative refractive indices that could have potential application in cloaking devices.7−9 A comprehensive understanding of the surface plasmon modes in discrete plasmonic structures, the underlying states of these photonic systems, is essential for both understanding and optimizing the design of these nanoscale elements. © 2016 American Chemical Society

Received: June 8, 2016 Accepted: August 2, 2016 Published: August 2, 2016 8552

DOI: 10.1021/acsnano.6b03796 ACS Nano 2016, 10, 8552−8563

Article

www.acsnano.org

Article

ACS Nano

Figure 1. (a) Schematic of the EELS geometry. (b,c) Comparison between the experimental results (b) and BEM simulations (c) of a gold trimer consisting of 45 nm diameter spheres and a 55° semiangle is shown. The inset in (b) shows an ADF STEM image of the trimer. The blue dots represent the recorded signal, averaged over a 3 × 3 pixel area (shown as the blue squares in the insets). The black spectrum represents the sum of the model components. The gray component extracted by NMF in (b) represents the ZLP tail. For simulated EELS, the model consisted of four Lorentzian resonances (α−δ) and the gold loss function (ε).

Figure 2. (a) Simulated eigenmode energies for systematic variation in the trimer semiangle as depicted in the overlaid schematic. Eigenmodes are defined in the quasi-static approximation. (b) (i−iv) Simulated (full BEM calculations with electrodynamic retardation effects) and experimental EELS maps of the plasmon modes detected in the nanosphere trimers, with corresponding ADF STEM and brightfield TEM images shown in (v). Scale bars are 50 nm. Gold nanospheres used in this work have diameters of approximately 45 nm. White numbers on the simulated EELS maps represent scaling factors. The simulated intensities for the 1.64 eV mode were low and are presented on a normalized intensity scale relative to the mode adjacent in energy (marked with a scaling factor of ×1). Greek letters correspond to NMF spectral factors corresponding to experimental maps (see also Figure 1 and Figure S2).

between two or more modes (e.g., Fano resonances) are consequently better understood by experimental measurements of both bright and dark plasmon modes.

that do not interact with light due to their negligible dipole moment (termed “dark modes”) cannot be easily probed optically. Optical effects that rely on coupling and interference 8553

DOI: 10.1021/acsnano.6b03796 ACS Nano 2016, 10, 8552−8563

Article

ACS Nano

plasmonic resonances of gold nanoparticle ensembles. Figure 1a displays a schematic of the experimental setup. The incident electron beam is scanned in a 2D raster over a region enclosing a nanoparticle cluster. At each point in the scan, the transmitted beam is collected by an annular dark-field (ADF) detector and an EELS spectrometer, resulting in the acquisition of an image and an EELS spectrum image data cube from the region. Figure 1b,c presents spectral analysis of EELS signals recorded from a trimer with a 55° semiangle between the terminal particles. This trimer has two mirror planes, one in the plane defined by the centers of each particle and a second mirror plane orthogonal to the first. The semiangle here is measured relative to the second orthogonal mirror plane, as additionally illustrated in Figure 2a. Figure 1b depicts electron energy loss from a selected area (3 × 3 pixels) recorded at the tip of the right terminal particle (blue square, inset). Non-negative matrix factorization (NMF) was applied to further decompose the entire spectrum image, and the separated spectral components are each presented according to their weighting at the selected trajectories.40,41 Briefly, NMF decomposes the spatio-spectral data cube into a linear combination of spectral factors and corresponding amplitude maps, indicating the weighting of each spectral factor at each pixel. The black line shows the sum of all components. The corresponding amplitude maps are shown in Figure 2b. The NMF decomposition model reproduces the original data with high fidelity and significantly reduced noise. The background due to the spread in the incident beam energies, the zero loss peak (ZLP) tail, was recovered by NMF decomposition and separated from the energy losses due to excitation of surface plasmon resonances of the trimer. The spectra were further separated into four features attributed to the surface plasmon resonances (α−δ) and one further component (ε) attributed to the interband transitions of gold. These were identified according to the dominant peak or band edge signature in the spectral factor, and the center or edge onset was used to estimate the energy of the resonance (see also Figure 2b for experimental energies). While the numerical NMF algorithm returns spectral factors with some residual satellite peaks, most factors are dominated by a single peak or show the shape of the interband edge and bulk plasmon excitations expected for gold. Here, the symmetry, and accordingly the spatial distribution of the EELS intensity, and the relative peak energies are the key points of comparison for understanding the underlying mode structure. NMF is well-suited to this analysis despite imperfect peak shape recovery as it imposes minimal assumptions (nonnegativity) on the data set and allows for separation of surface plasmon resonance features from an irregular background consisting of ZLP and bulk and interband transition contributions. Each of the components α−ε was reproduced in BEM simulations of an ideal gold trimer consisting of 45 nm diameter gold spheres arranged with a 55° semiangle. The BEM simulations were performed using a purely real dielectric environment with no substrate (Figure 1c; see also Figure 2 and Figure S1). The simulated EELS signal was also modeled as the sum of four spectral features (α−δ) and the gold loss function (ε), fitted using four Lorentzian peaks with the loss function derived from the dielectric function reported by Johnson and Christy.42 Minor discrepancies between the sum of the four Lorentzian components and the recorded signal in Figure 1c were observed at energies above 2.5 eV due to the truncation of the series of higher-order multipolar resonances

Scanning transmission electron microscopy−electron energy loss spectroscopy (STEM−EELS) is an effective technique for the interrogation of dark modes.17,18 The technique involves measuring the energy lost in a focused high-energy electron beam (80−300 keV) after interaction with a sample. When incident upon a metallic nanoparticle, a small amount of kinetic energy from the fast electrons can be transferred to the free electrons in the metal, leading to a collective plasmonic excitation of the electron gas and a characteristic energy loss in the transmitted beam.19−21 This interaction of the electron beam with nanoparticle surface plasmons, combined with the subnanometer resolving power of the STEM technique, enables spatially resolved mapping of electron energy losses22,23 and allows direct observation of plasmon modes that are inaccessible to optical techniques. 17 Furthermore, the STEM−EELS technique allows for the simultaneous morphological and spectral analysis of individual nanostructures, a significant advantage over current time-consuming methods used to correlate the optical scattering spectra with separate structural characterization by electron microscopy.24−28 The spatial information obtained through EELS enables definitive mode assignments excited in assembled plasmonic structures. Single metal nanoparticles including spheres,22,29,30 rods,22,31 cubes,32 bipyramids,33 and nanowires34−37as well as nanoparticle dimers,18,31,38 trigonal planar particle trimers,39 and nanosphere chains17 have been analyzed using STEM−EELS, but the technique has, to the best of our knowledge, yet to be applied for the systematic study of surface plasmon resonance modes in nonlinear 2D and 3D three-particle (trimer) and fourparticle (tetramer) assemblies. In this work, we present a comprehensive EELS investigation of the mode symmetries giving rise to surface plasmon excitations in trimers and tetramers. A complete analysis of the mode evolution of trimer structures upon ring-opening (i.e., moving from a three-fold symmetric trimer to a linear chain trimer) informed by boundary element method (BEM) simulations and analyzed on the basis of point group irreducible representations is presented and compared to experimental EELS data. The addition of a fourth particle promotes a greater range of possible self-assembled geometries in the form of 2D and unique 3D structures. We investigate the role of tetramer symmetry for both in-plane (2D) and out-of-plane (3D) geometries, drawing on numerical simulations and experimental EELS spectrum imaging. In turn, we report modal symmetry assignments and the mode symmetry and energy evolution as a function of trimer and tetramer geometry. The results from the present study are discussed with reference to the optical response of similar structures throughout.12 The combined systematic study of spatially resolved mode mapping and symmetry assignment by experimental EELS and eigenmode analysis enables correct assignment of mode contributions in these increasingly complex nanoparticle ensemble structures. These mode assignments are not possible directly from far-field spectroscopy or from simulations on a single structure, particularly due to the complex interplay and significant resonance energy shifts associated with geometric distortions. Our systematic mode symmetry analysis now outlines the optical design toolbox for three- and four-particle 2D and 3D nanophotonic structures.

RESULTS EELS measurements on trimer and tetramer systems offer simultaneous spectral and spatial information about the 8554

DOI: 10.1021/acsnano.6b03796 ACS Nano 2016, 10, 8552−8563

Article

ACS Nano of metal spheres43 not captured in the four lowest-energy peaks. The contribution of the gold bulk loss function is prominent in both the experimental and simulated cases. Furthermore, the relative energies of the four dominant contributions to the spectral features were reproduced in the simulation and in the experiment (see also Figure 2b). The relative intensity of the α component was lower in the experimental EELS than predicted by the simulated energy loss probability at a similar trajectory. However, substrate-, contamination-, and ligand-induced damping effects of lowenergy resonances44 and deviations from an ideal spherical geometry likely alter the relative intensities of each mode to a small degree. The experimental spectral components also appeared as broader peaks, consistent with higher damping as well as spectral broadening due to convolution with the ZLP. Spectral factors for each NMF analysis are presented in Figure S2. Figure 2 presents a series of eigenmode calculations (defined exclusively in the quasi-static approximation) as well as a set of experimental and simulated EELS maps (full electrodynamical BEM calculations). The overlaid particle geometry schematics show four representative structures, tracking the evolution of three-fold symmetric trimer (left) to a linear chain trimer (right) via bent intermediate geometries. Each of these structures has been defined by the semiangle between the two terminal particles. These angles are shown on each of the schematics in Figure 2a, and the trimers were analyzed on the basis of point group irreducible representations. In this instance, a semiangle of 30° corresponds to the symmetric trimer of the D3h point group; a 90° semiangle corresponds to a linear trimer of the D∞h point group, and intermediate structures belong to the C2v point group. The energy diagram plots the shifts in individual eigenmode energies as a result of the changing interparticle coupling for the geometry series. Surface charge diagrams for these mode evolutions are shown in the Supporting Information (Figure S1). Figure 2b(v) shows ADF STEM and bright-field TEM images of four gold nanoparticle trimers with semiangles of approximately 30, 35, 55, and 80° when moving from left to right, along with experimental and simulated EELS maps of these structures shown in Figure 2b(i−iv). Due to scan distortions during EELS acquisition, the ADF STEM and EELS maps were corrected by reference to the bright-field TEM images, giving rise to black regions where the images have been sheared, rotated, and cropped (see Methods). Critically, the EELS spectrum imaging results here allow for experimental mode assignment based on both spatial signal distributions as well as mode energies. Comparisons between simulations and experiments are common in far-field optical spectroscopy but do not allow for inspection of spatial signal distributions. In contrast, the direct comparisons between experimental EELS, BEM EELS simulations, and BEM eigenmode calculations presented here enable validation of mode symmetry assignments based on both spectral energies and spatial EELS signal distributions. These combined analyses are integral to the confident assignment of the observed surface plasmon modes identified by the experimental STEM−EELS mapping of trimers (Figure 2) and tetramers (Figure 4). A detailed description of the trends shown in Figure 2a is given in the Supporting Information. Briefly, the trends in individual mode energies arise from the extent of interparticle coupling in the near-field; as the terminal particles separate with increasing semiangle, reduced coupling results in changes to mode hybridization and an increase in many of the trimer mode

energies. In contrast, out-of-plane dipolar modes (e.g., B1 mode, Figure S1) exhibit minimal energy shifts as the in-plane semiangle does not alter hybridization of the individual particle modes significantly in this case. Figure 2b(i−iv) shows the EELS maps (both experimental and simulated) for the trimers shown in Figure 2b(v) that represent stages of this ringopening process. These maps are plotted on a normalized scale to accommodate comparisons between experimental and simulated maps across a range of signal intensities. The normalization factors are presented in the Supporting Information (see Table S11). EELS intensities are not directly interpretable in terms of optical properties;43 however, the underlying eigenmodes can be used to calculate optical responses to any incident electromagnetic field.45 The energies associated with the experimental EELS maps represent the energy of the dominant peak or band edge in the corresponding spectral factor (see Figure S2). The corresponding BEM simulations were performed in the fully retarded case rather than in the quasi-static approximation used for eigenmode simulations. An eigenmode decomposition of the retarded case has not been reported to date; however, the simulations accounting for retardation by considering the timevarying nature of the incident and induced fields provide the best possible model for comparison with experimental EELS. The resulting energies differ from the eigenmode energies due to retardation-induced red-shifting, but the underlying symmetries of the resonances are present in the simulations and in the experimental EELS maps. Beginning with the EELS maps shown in Figure 2b(i), the lowest-energy mode simulated for the D3h trimer, with an energy of 1.64 eV, shows EELS intensity around the perimeter of the structure, with the EELS intensity reaching a maximum at the furthest point on each nanosphere from the center of the structure. The ring-like displacement current is characteristic of a magnetic mode,1,46 and the symmetry of the mode was assigned to the A2′ irreducible representation in the D3h point group. The A2′ mode was not detected experimentally via EELS due to poor coupling to the incident electron beam; hence an experimental EELS map for this mode is not shown. BEM simulations revealed the A2′ mode to exhibit substantially weaker interaction with the electron beam as the simulated A2′ mode map is scaled by a factor of 12 in Figure 2b(i) relative to the maximum EELS signal in the simulated map for the 1.79 eV mode in Figure 2b(ii). This magnetic mode was experimentally observed in an EELS study of 25 nm diameter silver particle trimers,39 but the higher damping in gold due to its intrinsic dielectric properties,42 the size-dependent variation in EELS intensity of surface plasmon excitations in spheres,43 and the greater overlap with the ZLP tail for the 45 nm diameter gold particle trimers studied here preclude observation of this mode. However, by extending this investigation into a series of symmetry-broken trimers, the related B2 mode derived from this magnetic mode of the perfectly symmetric trimer was observed in these self-assembled gold trimers. As the semiangle of the trimer increases, and its symmetry alters to that of the C2v point group, the A2′ mode transforms into a B2 mode. This mode is evident for trimers with 35 and 55° semiangles, and the experimental and modeled EELS maps for these modes are shown in Figure 2b(i) with energies of 1.42 eV in the experimental data and 1.70 and 1.67 eV for the modeled maps. This discrepancy in energy (